developer-mesh

README

Developer Mesh - AI Agent Orchestration Platform

🚀 The production-ready platform for orchestrating multiple AI agents in your DevOps workflows

Connect AI models • Intelligent task routing • Real-time collaboration • Enterprise scale

🎯 Transform Your DevOps with AI Orchestration

DevOps teams struggle to integrate AI effectively - managing multiple models, coordinating agents, and optimizing costs. Developer Mesh solves this with intelligent orchestration that routes tasks to the right AI agent at the right time.

Why Developer Mesh?

• 🤖 Multi-Agent Orchestration: Register and coordinate multiple AI agents with different capabilities
• 🧠 Intelligent Task Assignment: Assignment engine routes tasks based on capability, performance, and cost
• ⚡ Real-time Collaboration: WebSocket-based coordination with binary protocol optimization
• 💰 Cost Optimization: Smart routing minimizes AI costs while maximizing performance
• 🏢 Enterprise Ready: Production AWS integration with circuit breakers and observability

🌟 Key Features

AI Agent Orchestration

• MCP Protocol Support: Full Model Context Protocol 2025-06-18 implementation (JSON-RPC 2.0)
• Connection Modes: Optimized for Claude Code, IDEs, and custom agents
• Capability-Based Discovery: Agents advertise their strengths (code analysis, security, documentation)
• Dynamic Load Balancing: Routes tasks to least-loaded agents in real-time
• Collaboration Support: Framework for agent coordination
• Workload Management: Track and optimize agent utilization

Security & Data Protection

• Per-Tenant Encryption: AES-256-GCM with unique keys per tenant (PBKDF2 key derivation)
• Credential Protection: All API keys and secrets encrypted at rest
• Forward Secrecy: Unique salt/nonce per encryption operation
• Authenticated Encryption: GCM mode prevents tampering and ensures data integrity
• Tenant Isolation: Cryptographic isolation prevents cross-tenant data access

Intelligent Task Assignment

• Multiple Assignment Strategies:
  • Round-robin: Distribute tasks evenly
  • Least-loaded: Route to agents with lowest workload
  • Capability-match: Match task requirements to agent strengths
  • Performance-based: Route to fastest agents
• Circuit Breakers: Automatic failover when agents fail
• Priority Queuing: Critical tasks get processed first

Multi-Tenant Embedding Model Management

• Enterprise-Grade Model Catalog:
  • Support for OpenAI, AWS Bedrock, Google, and Anthropic models
  • Dynamic model discovery with automatic catalog updates
  • Version tracking and deprecation management
• Per-Tenant Configuration:
  • Tenant-specific model access and quotas
  • Monthly and daily token limits with automatic enforcement
  • Custom rate limiting per model
  • Priority-based model selection
• Intelligent Model Selection:
  • Automatic selection based on task type and requirements
  • Cost-optimized routing with budget constraints
  • Quota-aware failover to alternative models
  • Circuit breaker pattern for provider resilience
• Comprehensive Cost Management:
  • Real-time cost tracking per tenant/model/agent
  • Budget alerts and automatic spending limits
  • Usage analytics and optimization recommendations
  • Detailed billing integration support

Real-time Communication

• MCP over WebSocket: Industry-standard Model Context Protocol (MCP) 2025-06-18
• JSON-RPC 2.0: Standard message format for all operations
• Connection Modes: Auto-detection for Claude Code, IDEs, and agents
• DevMesh Tools: All functionality exposed as standard MCP tools
• Resource Subscriptions: Real-time updates for workflows and tasks
• Binary Protocol Support: Optional compressed messages for efficiency
• Heartbeat Monitoring: Automatic reconnection handling

Standard Industry Tools Integration

• Pre-Built Provider Templates: GitHub, GitLab, Jira, Confluence, Harness.io ready to use
• Organization-Level Configuration: Each organization configures their own tool instances
• Tool Expansion for AI: Single provider expands into multiple granular MCP tools
• Production Resilience Patterns:
  • Circuit Breaker: Prevents cascading failures with automatic recovery
  • Bulkhead Pattern: Isolates failures to prevent system-wide impact
  • Request Coalescing: Deduplicates concurrent identical requests
  • Two-Tier Caching: L1 in-memory + L2 Redis for <10ms response times
• Permission-Based Filtering: Only expose operations the user's token can execute
• Comprehensive Metrics: Track execution time, cache hits, circuit breaker states

Dynamic Tool Integration with Advanced Operation Resolution

• Zero-Code Tool Addition: Add any DevOps tool without writing adapters
• Intelligent Discovery System:
  • Format Detection: Automatically detects OpenAPI, Swagger, custom JSON formats
  • Format Conversion: Converts non-OpenAPI formats to OpenAPI 3.0
  • Learning System: Learns from successful discoveries to improve future attempts
  • User-Guided Discovery: Accept hints to speed up discovery for non-standard APIs
• Advanced Operation Resolution:
  • Semantic Understanding: AI-powered matching of actions to operations
  • Self-Learning System: Improves resolution accuracy over time (15-20% gain)
  • Multi-Level Caching: Memory (L1) and Redis (L2) caching for <10ms resolution
  • Permission Filtering: Only shows operations users can execute
  • Resource Scoping: Handles namespace collisions intelligently
  • 95%+ Success Rate: For common operations across all tools
• Universal Authentication: OAuth2, API keys, bearer tokens, basic auth, custom headers
• Health Monitoring: Automatic health checks with configurable intervals
• Supported Tools: Any tool with an OpenAPI/Swagger specification can be integrated

📊 Use Cases

🎯 Intelligent Code Review

Route security reviews to specialized models, style checks to faster models

• Parallel analysis by multiple specialized agents
• Cost savings through intelligent routing
• Configurable routing strategies

📚 Multi-Agent Documentation

Coordinate multiple AI agents to generate comprehensive docs

• Different agents handle different sections
• Consistency through orchestration
• Automatic task distribution

🚨 Smart Incident Response

Route alerts to specialized agents based on severity and type

• Automatic escalation to appropriate agents
• Learning from resolution patterns
• Priority-based task queuing

🏗️ Architecture

System Architecture

graph TB
    subgraph "Client Layer"
        A1[AI Agents]
        A2[CLI Tools]
        A3[Web Dashboard]
    end
    
    subgraph "API Gateway"
        B1[MCP WebSocket Server<br/>:8080] <!-- MCP Protocol (JSON-RPC 2.0) -->
        B2[REST API<br/>:8081]
        B3[Auth Service]
    end
    
    subgraph "Core Services"
        C1[Task Router] <!-- Source: pkg/services/assignment_engine.go -->
        C2[Model Management]
        C3[Assignment Engine] <!-- Source: pkg/services/assignment_engine.go -->
        C4[Cost Tracker]
        C5[Dynamic Tools]
    end
    
    subgraph "Data Layer"
        D1[(PostgreSQL<br/>+ pgvector)]
        D2[(Redis<br/>Cache & Streams)] <!-- Source: pkg/redis/streams_client.go -->
        D3[S3 Storage]
    end
    
    subgraph "External Providers"
        E1[OpenAI]
        E2[AWS Bedrock]
        E3[Google AI]
        E4[Anthropic]
    end
    
    subgraph "Monitoring"
        F1[Prometheus]
        F2[Grafana]
        F3[Alert Manager]
    end
    
    A1 -->|Binary WS| B1
    A2 --> B2
    A3 --> B2
    B1 --> C1
    B2 --> B3
    B3 --> C2
    C1 --> C3
    C2 --> C4
    C2 --> E1
    C2 --> E2
    C2 --> E3
    C2 --> E4
    C1 --> D1
    C2 --> D1
    C2 --> D2
    C3 --> D1
    C5 --> D2
    B1 --> D3
    C2 --> F1
    F1 --> F2
    F1 --> F3

Embedding Model Management Architecture

graph LR
    subgraph "Model Catalog"
        MC[Model Catalog<br/>Repository]
        MD[Model Discovery<br/>Service]
        MV[Model Validator]
    end
    
    subgraph "Tenant Management"
        TM[Tenant Models<br/>Repository]
        TC[Tenant Config<br/>Manager]
        TQ[Quota Manager]
    end
    
    subgraph "Model Selection"
        MS[Model Selection<br/>Engine]
        CB[Circuit Breaker]
        FO[Failover Logic]
    end
    
    subgraph "Usage Tracking"
        UT[Usage Tracker]
        CT[Cost Calculator]
        BM[Billing Manager]
    end
    
    MD --> MC
    MV --> MC
    TC --> TM
    TQ --> TM
    MS --> MC
    MS --> TM
    MS --> CB
    CB --> FO
    UT --> CT
    CT --> BM
    MS --> UT

Core Components

• MCP Server: WebSocket server for real-time agent communication (apps/edge-mcp)
• REST API: HTTP API for tool management and integrations (apps/rest-api)
• Worker Service: Asynchronous task processing (apps/worker)
• Assignment Engine: Task distribution algorithms (pkg/services/assignment_engine.go)
• Vector Database: pgvector for semantic search and embeddings
• Event Queue: Redis Streams for asynchronous processing (pkg/redis/streams_client.go)

🚀 Quick Start

Prerequisites

• Go 1.24+ (workspace support)
• Docker & Docker Compose
• PostgreSQL 14+ with pgvector extension
• Redis 6.2+ (for streams support)
• AWS Account (optional - for Bedrock embeddings)

Option 1: Docker (Recommended)

# Clone repository
git clone https://github.com/developer-mesh/developer-mesh.git
cd developer-mesh

# Configure environment
cp .env.example .env
# Edit .env with your settings (AWS credentials optional for local dev)
# Note: Default encryption keys are provided for local dev only

# Start all services
docker-compose -f docker-compose.local.yml up -d

# Verify health
curl http://localhost:8080/health  # MCP WebSocket Server <!-- Source: pkg/models/websocket/binary.go -->
curl http://localhost:8081/health  # REST API Server

Option 2: Local Development

# Clone and setup
git clone https://github.com/developer-mesh/developer-mesh.git
cd developer-mesh

# Install dependencies
make deps  # Runs go mod tidy and go work sync

# Start infrastructure (PostgreSQL, Redis)
make dev-setup  # Creates .env if needed

# Run database migrations
make migrate-up

# Start services using Docker Compose
make dev  # Starts docker-compose.local.yml

# OR start services manually:
make run-edge-mcp  # Port 8080
make run-rest-api    # Port 8081
make run-worker      # Background worker

<!-- REMOVED: # Services will be available at: (unimplemented feature) -->
# MCP Server (WebSocket): http://localhost:8080 <!-- Source: pkg/models/websocket/binary.go -->
# REST API: http://localhost:8081

🎮 Usage Examples

Connect via MCP Protocol

Using websocat (Command Line)

# Standard MCP connection
wscat -c ws://localhost:8080/ws

# Initialize connection
> {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","clientInfo":{"name":"my-agent","version":"1.0.0"}}}

# List available tools
> {"jsonrpc":"2.0","id":2,"method":"tools/list"}

# Execute a DevMesh tool
> {"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"devmesh.task.create","arguments":{"title":"Test task"}}}

Claude Code Connection

# Connect as Claude Code client (optimized responses)
wscat -c ws://localhost:8080/ws -H "User-Agent: Claude-Code/1.0.0"

IDE Connection

# Connect from IDE (rich debugging info)
wscat -c ws://localhost:8080/ws -H "X-IDE-Name: VSCode"

Agent Connection (Go)

// MCP connection from Go application
headers := http.Header{
    "X-Agent-ID": []string{"security-agent"},
}
ws, _ := websocket.Dial("ws://localhost:8080/ws", "", "http://localhost")

// Initialize MCP session
msg := map[string]interface{}{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": map[string]interface{}{
        "protocolVersion": "2025-06-18",
        "clientInfo": map[string]interface{}{
            "name": "security-agent",
            "version": "1.0.0",
            "capabilities": []string{"security", "vulnerability-scan"},
        },
    },
}
websocket.JSON.Send(ws, msg)

Submit a Task via MCP

# Create a task using MCP tools/call
curl -X POST http://localhost:8080/ws \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 4,
    "method": "tools/call",
    "params": {
      "name": "devmesh.task.create",
      "arguments": {
        "title": "Security scan repository",
        "type": "security",
        "priority": "high"
      }
    }
  }'

# The MCP server coordinates task distribution based on agent capabilities
# Tasks are automatically assigned to the best available agent

Add a DevOps Tool

# Add GitHub to your DevOps tool arsenal
curl -X POST http://localhost:8081/api/v1/tools \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "github",
    "base_url": "https://api.github.com",
    "auth_type": "token",
    "credentials": {
      "token": "ghp_xxxxxxxxxxxx"
    }
  }'

# The system automatically discovers GitHub's capabilities
# and makes them available to your AI agents

Enhanced Tool Discovery Example

# Add a tool with non-standard API (e.g., SonarQube)
curl -X POST http://localhost:8081/api/v1/tools \
  -H "Authorization: Bearer $TOKEN" \
  -d '{
    "name": "sonarqube",
    "base_url": "https://sonar.example.com",
    "discovery_hints": {
      "api_format": "custom_json",
      "custom_paths": ["/api/webservices/list"],
      "auth_headers": {
        "Authorization": "Bearer squ_xxxxx"
      }
    }
  }'

# The discovery system will:
# 1. Try the custom path and detect it's custom JSON
# 2. Convert it to OpenAPI 3.0 format
# 3. Learn the pattern for future SonarQube instances
# 4. Make all endpoints available immediately

Monitor System Health

# Check MCP server health via HTTP
curl http://localhost:8080/health

# Check health via MCP resource
echo '{"jsonrpc":"2.0","id":1,"method":"resources/read","params":{"uri":"devmesh://system/health"}}' | \
  wscat -c ws://localhost:8080/ws

# Check REST API health
curl http://localhost:8081/health

# Metrics are exposed at /metrics endpoint (Prometheus format)
curl http://localhost:8080/metrics  # MCP Server metrics
curl http://localhost:8081/metrics  # REST API metrics

Available MCP Tools

Core DevMesh functionality exposed as MCP tools:

• devmesh.task.create - Create tasks
• devmesh.task.assign - Assign tasks to agents
• devmesh.task.status - Get task status
• devmesh.agent.assign - Assign work to agents
• devmesh.context.update - Update session context
• devmesh.context.get - Get current context
• devmesh.search.semantic - Semantic search
• Dynamic tool execution via registered tools

Available MCP Resources

Monitor system state via MCP resources:

• devmesh://system/health - System health status
• devmesh://agents/{tenant_id} - List of registered agents
• devmesh://tasks/{tenant_id} - Active tasks
• devmesh://context/{session_id} - Session context
• devmesh://tools/{tenant_id} - Available tools

📈 Performance Features

• Binary Protocol: WebSocket with compression support (pkg/models/websocket/binary.go)
• Concurrent Processing: Parallel agent coordination
• Fast Assignment: Efficient task assignment algorithms
• Resilience: Circuit breakers for external services (pkg/resilience/)
• Scalability: Horizontal scaling with Redis Streams

🛠️ Technology Stack

• Language: Go 1.24+ with workspace support
• Databases: PostgreSQL 14+ with pgvector extension, Redis 6.2+
• AI/ML: AWS Bedrock (primary), OpenAI, Google AI, Anthropic (via adapters)
• Message Queue: Redis Streams for webhook processing
• Storage: AWS S3 for context storage (optional)
• Protocol: MCP 2025-06-18 over WebSocket (JSON-RPC 2.0)
• Observability: Structured logging, Prometheus metrics

📚 Documentation

Getting Started

• Quick Start Guide
• Authentication Quick Start
• Embedding Quick Start
• Local Development

API Reference

• MCP Protocol Guide - Complete MCP implementation details
• REST API Reference
• MCP Server Reference
• Webhook API Reference
• Embedding API Reference
• Dynamic Tools API

Architecture

• System Overview
• Go Workspace Structure
• Multi-Agent Embedding Architecture
• Package Dependencies
• Agent Architecture

Developer Guides

• Development Environment
• Testing Guide
• Debugging Guide
• Authentication Implementation

Operations

• Configuration Guide
• API Key Management
• Security
• Operations Runbook
• Monitoring

Guides

• AI Agent Orchestration
• Multi-Agent Collaboration
• Cost Optimization
• Performance Tuning
• Production Deployment

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

Development Workflow

1. Fork the repository
2. Create a feature branch
3. Make your changes with tests
4. Run make pre-commit
5. Submit a pull request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• AWS Bedrock team for AI/ML infrastructure
• pgvector for vector similarity search
• OpenTelemetry for observability standards
• The Go community for excellent tooling

---